Thermal power generating unit deep peak regulation cost analysis method based on large language model

Abstract

The invention discloses a thermal power generating unit deep peak regulation cost analysis method based on a large language model. According to a question request input by a user, multi-source heterogeneous data collection is executed through an intelligent agent module driven by a large language model, and a webpage data intelligent capturing module driven by the large language model is adopted to analyze and extract coal market price, crude oil price and regional peak regulation compensation policy texts published on a target website in real time; a multi-agent cooperative computing framework is constructed, fuel cost, equipment life loss cost and peak regulation compensation income are calculated in parallel through a fuel consumption computing agent, an equipment life loss evaluation agent and a peak regulation compensation income agent, and total cost modeling under the deep peak regulation working condition is achieved; and finally, generating a user demand-oriented thermal power generating unit deep peak regulation cost analysis report. According to the method, multi-dimensional real-time analysis and comprehensive benefit evaluation of the thermal power peak regulation cost are realized through cooperative calculation of the intelligent agents, and the flexible regulation requirement of a novel power system is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of peak-shaving control and energy economic analysis of thermal power generating units, and relates to a deep peak-shaving cost analysis method for thermal power generating units based on large language model (LLM) and multi-agent collaborative computing. Background Art

[0002] As the penetration of renewable energy generation continues to increase, power grid operations face significant volatility and pressure to absorb renewable energy output. Against this backdrop, the deep peak-shaving capabilities of thermal power units have become a key regulatory tool for smoothing grid power fluctuations and ensuring safe and stable system operation. To participate in the deep peak-shaving ancillary services market, thermal power companies must have a comprehensive and detailed understanding of the costs incurred during the peak-shaving process for their own units in order to offer reasonable quotes in the peak-shaving market.

[0003] Peak-shaving cost calculation requires the integration of multimodal information, including real-time unit operating data, fuel market prices, and peak-shaving compensation policy documents. Existing systems utilize independent data processing modules, resulting in timing misalignment and a lack of semantic parsing. This inability to correlate with the time-specific provisions of peak-shaving compensation policies hinders real-time cost-benefit calculations. Peak-shaving compensation policies are frequently updated, and traditional methods rely on manual parsing of policy documents and configuration of model parameters, resulting in response delays of several hours to days. Summary of the Invention

[0004] The purpose of the present invention is to solve the problems of accurate cost quantification and decision-making lag in deep peak-shaving scenarios, and to provide a cost analysis method that integrates a large language model (LLM) and multi-agent collaborative computing. Through semantic parsing, data acquisition agents, data preprocessing and computing agents, deep peak-shaving cost analysis can be completed in real time and quickly.

[0005] The specific technical solutions provided by the present invention are as follows:

[0006] A method for analyzing the deep peak-shaving cost of thermal power units based on a large language model comprises the following steps:

[0007] S1. Obtain the user input requirement description according to the interactive interface;

[0008] S2. Based on semantic analysis, the Chain-of-Thought (CoT) technology is used to guide the large language model to generate structured task plans, clarifying the data collection scope and the computational model calling sequence;

[0009] S3, obtains task data through the data collection agent, obtains thermal power unit operation data through the OPC protocol, and calls the search engine API for the current day's coal trading price and local peak-shaving compensation policy;

[0010] S4. Clean and preprocess the acquired data through the data preprocessing agent;

[0011] S5. Utilize the cost calculation agent to calculate fuel costs, equipment losses, and compensation benefits;

[0012] S6. Based on the cost calculation results, a structured report on in-depth peak-shaving cost analysis is generated through a large language model.

[0013] Furthermore, in step S1, when the user inputs a requirement through the natural language interaction interface, the large language model is input together with the predetermined prompt.

[0014] Furthermore, in step S2, the ReAct mechanism is used to combine reasoning and action, and after each execution, observation is performed and then reasoning is performed to achieve COT task planning.

[0015] Furthermore, in step S3, multi-source data is obtained through SQL statements, search engine APIs, and local knowledge bases;

[0016] Furthermore, in step S4, a data cleaning and preprocessing tool is established, and the tool is called by the intelligent agent to fill the missing values ​​of the thermal power unit operation data to eliminate random fluctuations;

[0017] Furthermore, in step S5, a cost calculation model is established and provided to the agent for invocation, specifically including the following models:

[0018] S51. Fuel cost model. Fuel costs can be determined based on consumption during operation. Fuel includes coal and fuel oil required for low load. Coal consumption can be calculated using the counterbalance method: using boiler and turbine operating data to calculate boiler efficiency and turbine heat rate, and then calculate power generation coal consumption, thereby obtaining the coal cost during operation. The calculation formula is as follows:

[0019]

[0020] Among them, P is the unit load, t is the unit operation time, f1 is the standard coal price, b s is the standard coal consumption rate for power supply, b s The calculation formula is as follows:

[0021]

[0022] Among them, η p Take 99% for pipeline efficiency, ξ ap is the power consumption rate of the plant, q is the heat consumption rate of the steam turbine, η g is the boiler efficiency, and the steam turbine heat rate is calculated as follows:

[0023]

[0024] Among them, m i is the mass flow rate in and out of the system, Δh i is the enthalpy rise corresponding to the mass flow rate in and out of the system, P e is the generator power. The boiler efficiency calculation formula is as follows:

[0025]

[0026] Where D is the boiler evaporation capacity, h z is the steam enthalpy, h g is the feed water enthalpy, B is the fuel consumption, and Q1 is the low calorific value of the fuel.

[0027] The fuel cost calculation formula is as follows:

[0028] F2=C oi f oi

[0029] Among them, C oi is the fuel consumption, f oi For fuel prices.

[0030] S52, equipment loss cost model, uses the Manson-Coffin formula to calculate the relationship between the total strain amplitude of the rotor and the number of cracking cycles. The formula is:

[0031]

[0032] Among them, σ f ′ is the fatigue strength coefficient, ε f ′ is the fatigue ductility coefficient, N f is the number of cracking cycles, b and c are material indices. The formula for the additional loss cost of the unit's variable load operation is as follows:

[0033]

[0034] Among them, f un The cost of purchasing the machine.

[0035] S53. Economic benefit model. Economic benefits include electricity price benefits and peak load compensation benefits. The formula for calculating electricity price benefits is as follows:

[0036]

[0037] Where P(t) is the unit load at time t, t0 is the start time of deep peak regulation, and t3 is the end time of deep peak regulation.

[0038]

[0039] Among them, P wis the standard load of the unit under deep peak regulation, t1 is the time to reach deep peak regulation load, t2 is the time to end deep peak regulation load, and m is the gear quotation.

[0040] Furthermore, in step S6, a cost analysis report and a visual analysis chart are output according to user needs.

[0041] Compared with existing technologies, the present invention has the following advantages: By introducing a large language model (LLM) and a multi-agent collaborative computing architecture, it enables real-time collection and efficient analysis of multi-dimensional data such as fuel costs, equipment losses, and compensation benefits for thermal power units, improving the automation and response speed of data analysis. Through a web data intelligent crawling module, it automatically obtains and parses the latest regional peak-shaving compensation policy text, resolving the lag problem of traditional manual processing and improving the flexibility and accuracy of quotation strategies. It automatically generates user-specific cost analysis reports and supports chart output, facilitating a rapid understanding of the economic benefits of peak-shaving. Using a large language model-driven web data intelligent crawling module, it parses and extracts coal market prices, crude oil prices, and regional peak-shaving compensation policy texts posted on target websites in real time. A multi-agent collaborative computing architecture is constructed, using a fuel consumption calculation agent, an equipment life loss assessment agent, and a peak-shaving compensation benefit agent to concurrently calculate fuel costs, equipment life loss costs, and peak-shaving compensation benefits, achieving total cost modeling under deep peak-shaving conditions. Ultimately, a user-specific deep peak-shaving cost analysis report for thermal power units is generated. Through intelligent collaborative computing, the present invention realizes multi-dimensional real-time analysis and comprehensive benefit evaluation of thermal power peak-shaving costs, effectively improves the economy and operation safety of units in deep peak-shaving scenarios, and meets the flexible adjustment needs of new power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flowchart of the steps provided by the present invention.

[0043] Figure 2 This is the specific method flow for intelligently decomposing and completing complex tasks based on a large language model provided by the present invention. DETAILED DESCRIPTION

[0044] To make the objectives, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0045] A specific embodiment provides a method for analyzing the deep peak load of thermal power units based on a large language model. Figure 1 As shown, the following steps are included:

[0046] S1. Obtain the user input requirement description according to the interactive interface;

[0047] S2. Based on semantic analysis, the thought chain technology is used to guide the large language model to generate structured task plans, clarifying the data collection scope and the calculation model calling sequence;

[0048] S3, obtains task data through the data collection agent, obtains thermal power unit operation data through the OPC protocol, and calls the search engine API for the current day's coal trading price and local peak-shaving compensation policy;

[0049] S4. Clean and preprocess the acquired data through the data preprocessing agent;

[0050] S5. Utilize the cost calculation agent to calculate fuel costs, environmental costs, equipment losses, and compensation benefits;

[0051] S6. Based on the cost calculation results, a structured report on in-depth peak-shaving cost analysis is generated through a large language model.

[0052] Specifically, in this example, in step S1, the user can manually input natural language instructions or give instructions through voice, and the language-to-text conversion is achieved through the existing TTS (speech synthesis) tool. For example, the user inputs "Evaluate the peak-shaving economy of 45% of the daily load", and the following steps are performed according to the current problem.

[0053] Furthermore, in this embodiment, in step S2, the method of guiding the large language model to generate a structured task plan by using the thought chain technology to clarify the data collection scope and the calculation model calling sequence includes:

[0054] S21. Based on the reasoning and action coordination mechanism of the ReAct framework, combined with the step decomposition capability of the Chain of Thought (CoT), a structured task plan is generated.

[0055] Specifically, the entire framework includes a task description, tool description, answer format, final result, and answer process. The task description is expressed as a cost analysis of deep peak-shaving for thermal power units. Tools include data collection tools, data preprocessing tools, and calculation tools. In addition, a function call interface is required to provide a large language model for invocation. The answer format is based on the user question, the answer strategy, the action to be performed (i.e., the tool to be called), the parameters input to the calling tool, and the result returned by the action. For example, "You are an expert in deep peak-shaving cost analysis for thermal power units. Users will input different questions. You need to analyze and answer user questions based on the following tools and dialogues. You can use the following tool {tool}. When answering questions, please use the following format to analyze and answer the questions step by step: Question: the question you need to answer, Thought: the idea of ​​answering the question, Action: the action taken, which should be one of the following tools: {tool}, Action Input: the input of the action taken, Observation: the result returned by the action, (Thought / Action / ActionInput / Observation can be repeated many times), Thought: I now know the final answer, Final Answer: the final answer to the original input question. The previous dialogue message history is: {history}." Among them, the content not enclosed in brackets is fixed content, and the content in brackets is replaceable content. According to the specific method flow of intelligent decomposition and completion of complex tasks based on large language models (such as Figure 2 ).

[0056] S22. Data collection scope and calculation model calling.

[0057] Specifically, the large language model collects data and calls models based on different questions input by users. According to the question "Evaluate the economic feasibility of peak-shaving with 45% load on that day", the large language model will guide the system to prioritize obtaining unit operation data, today's coal price, knowledge base data and local compensation policies.

[0058] Furthermore, in this embodiment, in step S3, multi-source data is acquired through overall data collection.

[0059] S31. Based on the problem, use the tool to call SQL statements to obtain operating data, such as main steam flow, main steam enthalpy, reheat steam flow, reheat steam enthalpy, rated power, high-pressure cylinder exhaust enthalpy, boiler evaporation rate, feed water enthalpy, steam enthalpy, low heating value of coal, and the deep peak regulation operation time of the unit on that day.

[0060] S32. Based on the question, obtain the fuel price and peak-shaving compensation policy text for the day through the tool search engine API.

[0061] S33. Based on the question, retrieve the fatigue strength coefficient, elastic modulus, fatigue strength coefficient, fatigue plasticity coefficient, fatigue plasticity index, plant power consumption rate, machine purchase cost and fuel consumption of the current unit rotor material through the knowledge base.

[0062] Furthermore, in this embodiment, in step S4, the acquired data is cleaned and preprocessed by a data preprocessing agent;

[0063] S41 and the large language model perform data cleaning and preprocessing based on the acquired data. The missing value filling tool uses the K nearest neighbor algorithm and the sliding window mean method to eliminate random fluctuations.

[0064] Furthermore, in this embodiment, in step S5, the cost calculation agent is used to calculate the economic benefits.

[0065] S51. The fuel cost is calculated by the fuel calculation agent based on the preprocessed data. The large language model calls the fuel calculation tool, which encapsulates the function according to the calculation formula of the fuel cost. The calculation formula is as follows:

[0066]

[0067] Among them, P is the unit load, t is the unit operation time, f1 is the standard coal price, b s is the standard coal consumption rate for power supply, b s The calculation formula is as follows:

[0068]

[0069] Among them, η p Take 99% for pipeline efficiency, ξ ap is the power consumption rate of the plant, q is the heat consumption rate of the steam turbine, η g is the boiler efficiency, and the steam turbine heat rate is calculated as follows:

[0070]

[0071] Among them, m i is the mass flow rate in and out of the system, Δh i is the enthalpy rise corresponding to the mass flow rate in and out of the system, P e is the generator power. The boiler efficiency calculation formula is as follows:

[0072]

[0073] Where D is the boiler evaporation capacity, h z is the steam enthalpy, h g is the feed water enthalpy, B is the fuel consumption, and Q1 is the low calorific value of the fuel.

[0074] The fuel cost calculation formula is as follows:

[0075] F2=C oi f oi

[0076] Among them, C oi is the fuel consumption, f oi For fuel prices.

[0077] S52. The large language model calls the equipment loss calculation tool. The tool encapsulates functions according to the equipment loss cost calculation formula. The calculation formula is as follows:

[0078]

[0079] Among them, σ f ′ is the fatigue strength coefficient, ε f ′ is the fatigue ductility coefficient, N f is the number of cracking cycles, b and c are material indices. The formula for the additional loss cost of the unit's variable load operation is as follows:

[0080]

[0081] Among them, f un The cost of purchasing the machine.

[0082] S52. The large language model calls the economic benefit calculation tool. The tool encapsulates functions according to the economic benefit cost calculation formula. The electricity price benefit calculation formula is as follows:

[0083]

[0084] Where P(t) is the unit load at time t, t0 is the start time of deep peak regulation, and t3 is the end time of deep peak regulation.

[0085] The formula for deep peak load regulation compensation income is as follows:

[0086]

[0087] Among them, P w is the standard load of the unit under deep peak regulation, t1 is the time to reach deep peak regulation load, t2 is the time to end deep peak regulation load, and m is the gear quotation.

[0088] Furthermore, in this embodiment, in step S6, a structured report of in-depth peak-shaving cost analysis is generated through a large language model.

[0089] S61. The cost calculation tool is called based on the large language model in step 5 to obtain the calculation results, and a cost analysis report is generated based on the results. For example, based on the question "Evaluate the economic feasibility of peak load regulation at 45% of the current day's load," the LLM analysis results are as follows: For the economic evaluation of peak load regulation at 45% for a 660MW unit, the key parameters are estimated as shown in Table 1:

[0090] Table 1 Economic indicators under 45% load

[0091] parameter 50% load (330MW) 45% load (297MW) Power supply coal consumption (g / (kW·h)) 311.53 329.16 Factory power consumption rate (%) 8.4％ 9.15％ Boiler thermal efficiency (%) 90.39％ 89.89％ Steam turbine heat rate (kJ / (kW·h)) 8590 8905

[0092] The peak load regulation duration is 6 hours, the unit price of standard coal is 1540 yuan / ton, and the on-grid electricity price is 0.391 yuan / (kW·h). The peak load compensation rules are as follows: 50% to 40% load: 0.6 yuan / (kW·h), 40% to 30% load: 1.0 yuan / (kW·h). 45% load (297MW) is in the 50% to 40% range, and the peak load compensation is calculated at 0.6 yuan / (kW·h). Adjusting from full load (660MW) to 45% (297MW) is a 55% change. The corresponding life loss rate of 55% load change is 0.0013. The cost and benefit analysis is shown in Table 2:

[0093] Table 2 Cost and benefit analysis

[0094] project 45% load peak shaving 50% base load Power generation income (10,000 yuan) 297×6×0.391＝698.5 330×6×0.391＝775.9 Peak load compensation (10,000 yuan) (330-297)×6×0.6＝118.8 0 Total revenue (10,000 yuan) 817.3 775.9 Fuel cost (10,000 yuan) <![CDATA[297×6×0.32916×1540 / 10 6 =91.2]]> <![CDATA[330×6×0.31153×1540 / 10 6 =93.5]]> Equipment loss cost (10,000 yuan) 4.695 0 Net income (10,000 yuan) 726.1 682.4

[0095] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for analyzing the deep peak load regulation cost of thermal power units based on a large language model, characterized in that: The steps include: S1. Obtain the user input requirement description according to the interactive interface; S2. Based on semantic analysis, the Chain-of-Thought (CoT) technology is used to guide the large language model to generate structured task plans, clarifying the data collection scope and the computational model calling sequence; S3, obtains task data through the data collection agent, obtains thermal power unit operation data through the OPC protocol, and calls the search engine API for the current day's coal trading price and local peak-shaving compensation policy; S4. Clean and preprocess the acquired data through the data preprocessing agent; S5. Utilize the cost calculation agent to calculate fuel costs, equipment losses, and compensation benefits; S6. Based on the cost calculation results, a structured report on in-depth peak-shaving cost analysis is generated through a large language model.

2. The method for analyzing the deep peak-shaving cost of thermal power units based on a large language model according to claim 1, characterized in that: In step S1, when the user inputs a requirement through the natural language interaction interface, the user inputs the large language model together with the predetermined prompt.

3. The method for analyzing the deep peak-shaving cost of thermal power units based on a large language model according to claim 1, characterized in that: In step S2, the ReAct mechanism is used to combine reasoning and action. After each execution, observation is performed and then reasoning is performed to achieve COT task planning.

4. The method for analyzing the deep peak-shaving cost of thermal power units based on a large language model according to claim 1, characterized in that: In step S3, a data cleaning and preprocessing tool is established, and the tool is called by the intelligent agent to fill the missing values ​​of the thermal power unit operation data to eliminate random fluctuations.

5. The method for analyzing the deep peak-shaving cost of thermal power units based on a large language model according to claim 1, characterized in that: In step S4, a cost calculation model is established and provided to the agent for invocation, specifically including the following models: S41. Fuel cost model. Fuel costs can be determined based on consumption during operation. Fuel includes coal and fuel oil required for low load. Coal consumption can be calculated using the counter-balance method: boiler efficiency and turbine heat rate are calculated using boiler and turbine operating data, and power generation coal consumption is calculated to obtain the coal cost during operation. The calculation formula is as follows: Among them, P is the unit load, t is the unit operation time, f1 is the standard coal price, b s is the standard coal consumption rate for power supply, b s The calculation formula is as follows: Among them, η p Take 99% for pipeline efficiency, ξ ap is the power consumption rate of the plant, q is the heat consumption rate of the steam turbine, η g is the boiler efficiency, and the steam turbine heat rate is calculated as follows: Among them, m i is the mass flow rate in and out of the system, Δh i is the enthalpy rise corresponding to the mass flow rate in and out of the system, P e is the generator power, and the boiler efficiency is calculated as follows: Where D is the boiler evaporation capacity, h z is the steam enthalpy, h g is the feed water enthalpy, B is the fuel consumption, Q1 is the low calorific value of the fuel, The fuel cost calculation formula is as follows: F2=C oi f oi Among them, C oi is the fuel consumption, f oi is the fuel price; S42, equipment loss cost model, uses the Manson-Coffin formula to calculate the relationship between the total strain amplitude of the rotor and the number of cracking cycles. The formula is: Among them, σ f ′ is the fatigue strength coefficient, ε f ′ is the fatigue ductility coefficient, N f is the number of cracking cycles, b and c are material indexes, and the formula for the additional loss cost of the unit's variable load operation is as follows: Among them, f un For the purchase cost, S43. Economic benefit model. Economic benefits include electricity price benefits and peak load compensation benefits. The formula for calculating electricity price benefits is as follows: Where P(t) is the unit load at time t, t0 is the start time of deep peak regulation, and t3 is the end time of deep peak regulation. Among them, P w is the standard load of the unit under deep peak regulation, t1 is the time to reach deep peak regulation load, t2 is the time to end deep peak regulation load, and m is the gear quotation.

6. The method for analyzing the deep peak-shaving cost of thermal power units based on a large language model according to claim 1, characterized in that: In step S5, a cost analysis report and a visual analysis chart are output according to user requirements.

Images